application of taguchi method in health and safety 2/h45023744.pdf · application of taguchi method...
TRANSCRIPT
Alharthi A. A et al Int. Journal of Engineering Research and Applications www.ijera.com
ISSN : 2248-9622, Vol. 4, Issue 5( Version 2), May 2014, pp.37-44
www.ijera.com 37 | P a g e
Application of Taguchi Method In Health And Safety (Fire
Extinguishing Experiment)
Alharthi A. A. & Yang Q. School of Engineering , Brunel University, London
ABSTRACT The traditional Taguchi method is widely used for optimizing the process parameters of a single response
problem. In this paper, Taguchi method is applied to study the effects of five control variables – training,
experience, response to alarm, age, and qualification on extinguishing time and percent damage. An L16
orthogonal array (OA) was used to accommodate the experiment. ANOVA and F-tests and regression are used
to analyze the results. The study indicated that training and experience have the largest effect on the on
extinguishing time and percent damage
Key words: ANOVA analyze, Extinguishing time, Fire extinguishing experiment, orthogonal array (OA),
percent damage and Taguchi method.
I. Introduction Taguchi Method or Robust Engineering,
developed by Genichi Taguchi, is an approach to
Design of Experiments (DOE) for designing products
or processes so that they are robust to environmental
conditions such temperature and humidity. The
objective of Taguchi method is to model responses
(and variance) as a function of controllable (and
uncontrollable) factor levels, then choose levels of
controllable factors to reduce variation transmitted to
the response from variation of the controllable factors
and of the uncontrollable factors, in another word
reduce product variation by choosing levels of the
control factors that dampen the effect of the
uncontrollable or noise factors. Quality is improved
without controlling or removing the cause of
variation, instead, we make the product (or process)
robust to variation in the noise factors [4], [5], [6].
Noise factor is measured by signal to noise ratio and
it is calculated depends on the objective of the
experiment. There are three ways the response could
be optimized[6]:
Objective Signal to noise ratio
Minimize response -10*log(Sy2/n)
(smaller the better)
Maximize the response -10*log(S(1/y2)
/n)
(larger the better)
Nominal is best -10*log(s2)
Multiplied by 10 to put into “deci” “bel” metric, a
terminology used in Electrical Engineering. Taguchi
suggested that “quality” should be thought of, not as
a product being inside or outside of specifications,
but as the variation from the target. He defines
quality as the losses a product imparts to the society
from the time the product is shipped. To quantify
quality loss, write T for the target value and Y for the
measured value [5], [6]. We want E (Y) = T. Write
L(Y) for the loss (in dollars, reputation, customer
satisfaction, ……) for deviation of Y from T. The loss
function is L(Y) = K(Y-T)2
Where, K is some constant. If E (Y) really is T, then
E(L(Y)) = Kσ2 , where σ
2 = Var (Y).
If the product is off target, so that E (Y) = T +d, then
E(L(Y )) = k(σ2+d
2).
Figure 1. Quality loss function
A Taguchi design, or an orthogonal array, is
a method of designing experiments that usually
requires only a fraction of the full factorial
combinations. An orthogonal array means the design
is balanced so that factor levels are weighted equally.
Because of the orthogonality, each factor can be
evaluated independently of all the other factors, so
RESEARCH ARTICLE OPEN ACCESS
Alharthi A. A et al Int. Journal of Engineering Research and Applications www.ijera.com
ISSN : 2248-9622, Vol. 4, Issue 5( Version 2), May 2014, pp.37-44
www.ijera.com 38 | P a g e
the effect of one factor does not influence the
estimation of another factor.
The Steps followed for Taguchi design are:
1. State your problem and objective
2. List responses, control parameters, and sources
of noise
3. Plan the experiment
4. Run experiment and
5. Analyze the experiment and predict improved
parameter settings
II. Problem Identification Fire accidents are common and play an
important role amongst major accidents, not only
because of their relatively high frequency but also
because of their dangerous effects. This makes the
control and protection from the fire accidents a vital
issue that needs to be studied. Our objective is to
determine the effect of employee training,
experience, his or her response to alarm, his or her
age and qualification and their interactions on how
long it takes to extinguish a fire and asses time the
percent damage resulted from the fire.
III. Experimental Details A series of experimental tests were designed
to accomplish these objectives and develop baseline
for future research. The experiment is conducted by
running it at various levels of the factors.
3.1 Determining factors for the study
It was determined the human factors that
need to be studies which influence the performance
of an employee in using the extinguisher are:
a) Training: it is expected that there is difference
between the performance of trained employee
and untrained employee.
b) Experience: the employee with more experience
may perform better in the extinguishing process.
It is worthy to say that the expert employee is
already a trained employee.
c) The response to the alarm: the fast response to
the alarm may lead us to the best scenarios.
d) Age: the employee age directly affects the
physical and mental behavior of the employee in
which it affects his performance.
e) Qualification: higher qualification is predicted to
lead to better performance.
The response variables to be measured and
improved are:
a) Extinguishing time: it is a measure of the
employee performance and measured in seconds.
b) Percentage of damage: it will be used to study
the relation between the extinguishing time and
the damage percentage. It is believed longer
extinguisher time leads to more damage.
3.2 Determining the levels of the factors The Process parameters and their levels used
in the experiment summarized in table 1
Table 1. Factors and their levels
Factors code Level 0 Level 1
Training T Untrained Trained
Experience X without
experience
with
experience
Response to
the alarm
R >30 sec.
(Slow
response)
<30 sec.
(Fast
response)
Age G >40 <40 years
Qualification Q <Bachelor >Bachelor
3.3 Taguchi Design
Taguchi Orthogonal Array L16 design is
used (table 2) which included 5 factors and 16 runs.
Columns of L16(2**15) Array are chosen that are 1,
2, 4, 8 and 15. The L16 is a resolution III design
which means the main effect is confounding with two
factor interaction. The alias structure for L16 is
summarized below:
[A] = A - BC - DE - FG - HJ -KL - MN - OP
[B] = B - AC - DF - EG - HK -JL - MO - NP
[C] = C - AB - DG - EF - HL - JK - MP - NO
[D] = D - AE - BF - CG - HM - JN - KO - LP
[E] = E - AD - BG - CF - HN - JM - KP - LO
[F] = F - AG - BD - CE - HO - JP - KM - LN
[G] = G - AF - BE - CD - HP - JO - KN - LM
[H] = H - AJ - BK - CL - DM - EN - FO - GP
[J] = J - AH - BL - CK - DN - EM - FP - GO
[K] = K - AL - BH - CJ - DO - EP - FM – GN
[L] = L - AK - BJ - CH - DP - EO - FN - GM
[M] = M - AN - BO - CP - DH - EJ - FK – GL
[N] = N - AM -BP - CO - DJ - EH - FL – GK
[O] = O - AP - BM - CN - DK - EL - FH - GJ
[P] = P - AO - BN - CM - DL -EK - FJ - GH
Based on the alias structure, factor C is
confounded with AB interaction, Factor E is
confounded with AD interaction, factor F is
confounded with BD interaction and so on.
The first linear graph for L16 (figure 2)
assisted in matching factors and column and possible
interaction in the experimental matrix [3].
Figure 2. First Linear Graph for L16 Array assigns
the variables and interactions
Alharthi A. A et al Int. Journal of Engineering Research and Applications www.ijera.com
ISSN : 2248-9622, Vol. 4, Issue 5( Version 2), May 2014, pp.37-44
www.ijera.com 39 | P a g e
No replicates of treatment combinations were done,
also the experiment was run in random order to
minimize the effect of extraneous variables that
might influence the results. Alpha (α) or type I error
of 0.05 is used. Alpha is the maximum acceptable
level of risk for rejecting a true null hypothesis.
Run # Factors Extinguishing
Time
Damage
percentage
T X R G Q (seconds) %
1 2 4 8 15
1 0 0 0 0 0 253 84
2 0 0 0 1 1 229 82
3 0 0 1 0 1 219 75
4 0 0 1 1 0 171 73
5 0 1 0 0 1 150 52
6 0 1 0 1 0 125 49
7 0 1 1 0 0 115 18
8 0 1 1 1 1 107 16
9 1 0 0 0 1 160 51
10 1 0 0 1 0 120 26
11 1 0 1 0 0 116 15
12 1 0 1 1 1 96 12
13 1 1 0 0 0 110 20
14 1 1 0 1 1 108 13
15 1 1 1 0 1 96 12
16 1 1 1 1 0 92 8
Table 2 L16 orthogonal array design matrix and the results
Alharthi A. A et al Int. Journal of Engineering Research and Applications www.ijera.com
ISSN : 2248-9622, Vol. 4, Issue 5( Version 2), May 2014, pp.37-44
www.ijera.com 40 | P a g e
3.4 Equipment
Extinguishers: Class A&B extinguishers will
be used. (Foam Extinguisher)
The burned material: Crib of wooden sticks
will be used as a burned material.
Tray containing heptane to light the fire.
Personal protective equipment.
3.5 Experiment Assumptions
The experiments are conducted in the same
conditions, which mean that the place; the burning
material and the extinguishing method are the same
in all the trials. In addition to that, the time for
starting the extinguishing is the same.
3.6 Experiment Conditions
The test room where the experiment was
conducted was closed with the exception of a small
opening at the base of the door (Provided for
ventilation). The wood cribs description is shown in
Table 3.
Table 3: Wood Description
CLASS DIMENSIONS (m)
White Wood 0.5*0.5*0.5
The wood cribs are fixed at 50 cm above
floor level. A properly sized tray is placed beneath
the crib at 30 faraway from the wood cribs (see
Figure 3 and Figure 4). The appropriate heptane
starter charge is poured into the tray. The heptane
charge is ignited and allowed to ignite the wood crib
above [2]. The wood crib is allowed to burn for a
period of 30 seconds before extinguishing (see Figure
5).
For each run, the following steps should be
followed:
Put the wood crib at the center of the
experiment room, and place a tray full of
heptane under it to light the fire.
Ignite the heptane.
Start the extinguishing process after 30
seconds from removing the tray.
Write down the time spent in the
extinguishing process, and the percentage of
damage in the wood for each trial.
IV. Experimental Analysis and Discussion 4.1 Graphical Analysis
4.1.1 Graphical Analysis for Extinguishing Time
The main effects plot for extinguishing time
(figure 6) shows that extinguishing time decreases for
a trained and experienced employee, also it decreases
for a response to alarm for less than 30 seconds,
concluding that training, experience and response to
alarm variables are significant factors while age and
qualification factors are not as significant.
Qualification could be ignored and it may not be used
as factor to improve extinguishing time.
Alharthi A. A et al Int. Journal of Engineering Research and Applications www.ijera.com
ISSN : 2248-9622, Vol. 4, Issue 5( Version 2), May 2014, pp.37-44
www.ijera.com 41 | P a g e
10
168
156
144
132
120
10 10
10
168
156
144
132
120
10
Training
Me
an
Experience Response to Alarm
Age Qualification
Main Effects Plot for Extinguishing TimeData Means
Figure 6. The main effect plot for extinguisher time.
The interactions plot for extinguishing time
(figure 7) shows that there is a significant interaction
between training and experience. The plot indicates
that there is no other interaction exists. We will only
include the Training * Experience interaction as a
term in the statistical analysis.
10 10 10 10
200
150
100
200
150
100
200
150
100
200
150
100
Training
Experience
Response to Alarm
Age
Qualification
0
1
Training
0
1
Experience
0
1
to Alarm
Response
0
1
Age
Interaction Plot for Extinguishing TimeData Means
Figure 7. The interaction plot for extinguisher time
4.1.2 Graphical Analysis for Percent Damage
The main effects plot for the percent damage
(figure 8) shows % damage decreases for a trained
and experienced employee, also it decreases for a
response to alarm for less than 30 seconds,
concluding that training, experience and response to
alarm variables are significant factors while age and
qualification factors are not statistically significant.
10
60
50
40
30
20
10 10
10
60
50
40
30
20
10
Training
Me
an
Experience Response to Alarm
Age Qualification
Main Effects Plot for % DamageData Means
Figure 8. Main Effects Plot for Percent Damage.
The interaction plot (figure 9) shows that
there is a significant interaction between training and
experience, and between age and qualification. We
will include the training * experience and age
qualification interactions as terms in the statistical
analysis.
10 10 10 10
70
45
20
70
45
20
70
45
20
70
45
20
Training
Experience
Response to Alarm
Age
Qualification
0
1
Training
0
1
Experience
0
1
to Alarm
Response
0
1
Age
Interaction Plot for % DamageData Means
Figure 9. Interactions plot for percent damage
4.2 Statistical Analysis
ANOVA or analysis of variance and
multiple regression statistical methods were used to
analyze the data generated by our experiment.
ANOVA is useful for determining the influence of
any giving input parameter from a series of
experimental results for the fire experiment. In
general, ANOVA compares the variation between
groups and the variation within samples by analyzing
their variances. It partitions the total variation into its
appropriate components[1].
Total variance = between groups variance +
variance due to the errors
Where SST = Total Sum of Squares; SSG =
Treatment Sum of Squares between the groups; SSE
= Sum of Squares of Errors. Just think of 'sums of
squares' as being a measure of variation. The method
of measuring this variation is variance, which is
standard deviation squared.
There are 3 assumptions for the ANOVA Statistical
F-test to be valid which involve the εij’s (the error
terms) and are summarized below:
1. The εij’s are normally distributed.
2. The εij’s have mean zero and a common
variance, σ2.
3. The εij’s are independent across
observations.
Similar to Analysis of variance (ANOVA),
multiple linear regression is used to model the
relationship between response variables and one or
more independent variables and variables. The βi are
the regression parameters and ε is an error. The least
square regression method is used for fitting model.
Similar to Analysis of variance (ANOVA), multiple
linear regression is used to model the relationship
between response variables and one or more
Alharthi A. A et al Int. Journal of Engineering Research and Applications www.ijera.com
ISSN : 2248-9622, Vol. 4, Issue 5( Version 2), May 2014, pp.37-44
www.ijera.com 42 | P a g e
independent variables and their relevant interactions
[1]. The model for the multiple regression equation
is: Y = β0 + β1X1 + β2X2 + β3X3 + β4X4 + ………βnXn +
ε
Where y is the response or the dependent variable
and are the independent X1, X2, X3, X4 ...........and Xn
are independent variables. The βi are the regression
parameters and ε is an error. The least square
regression method is used for fitting model.
4.2.1 Statistical Analysis for Extinguishing Time
Statistical outputs for extinguishing time are
summarized in table 4 and 5. Regression analysis
provides the coefficients for the factors and their p-
values and an analysis of variance table. The order of
the coefficients by absolute value indicates the
relative importance of each factor to the response; the
factor with the biggest coefficient has the greatest
impact. The sequential sums of squares in the
analysis of variance table also indicate the relative
importance of each factor; the factor with the biggest
sum of squares has the greatest impact.
Table 4. Regression analysis for Extinguishing
Time .
The regression equation is:
Extinguishing Time = 240 - 95.0 Training - 93.8
Experience - 30.4 Response to Alarm - 21.4 Age +
7.87 Qualification + 72.3 T*X
Predictor Coef SE Coef T P
Constant 239.938 7.725 31.06 0.000
Training -95.000 8.258 -11.50 0.000
Experience -93.750 8.258 -11.35 0.000
Response to Alarm -30.375 5.839 -5.20 0.001
Age -21.375 5.839 -3.66 0.005
Qualification 7.875 5.839 1.35 0.210
T*X 72.25 11.68 6.19 0.000
S = 11.6789 R-Sq = 96.9% R-Sq(adj) = 94.8%
Table 5. Analysis of Variance for Extinguishing
Time
Analysis of Variance
Source DF SS MS F P
Regression 6 38133.9 6355.6 46.60 0.000
Residual Error 9 1227.6 136.4
Total 15 39361.4
Source DF Seq SS
Training 1 13865.1
Experience 1 13282.6
Response to Alarm 1 3690.6
Age 1 1827.6
Qualification 1 248.1
T*X 1 5220.1
The regression analysis shows that the P-
value for Training, Experience, Response to Alarm,
Age, and the Training*Experience interaction are 0,
0, 0.001, 0.005, and 0 respectively which are much
smaller than Alpha of 0.05, indicating statistical
significance, while the P-value for Qualification is
0.21 which is greater than Alpha of 0.05, indicating
non statistical significance.
The prediction equation is:
YTime = 240 – 95XT – 93.8XX – 30.4XR – 21.4XG+
7.87 XQ + 72.3XTXX… (Equation (1))
The residual plot (figure 10) indicates that there is no
violation of the analysis of variance assumptions.
The residuals are normally distributed, the residuals
have equal variances, and the residual are
independent. This concludes that our model is valid.
4.2.2 Statistical Analysis for Percent Damage Statistical outputs for percent damage are
summarized in table 6 and 7. The regression analysis
shows that the P-value for training, experience,
response to alarm, training*experience interaction,
and age*qualification interaction are 0, 0, 0, 0, and
0.004 respectively which are much smaller than
Alpha of 0.05, indicating statistical significance,
while the P-value for age, and qualification are 0.056
and 0.379 respectively which are greater than Alpha
of 0.05, indicating non statistical significance.
Table 6. Regression analysis for % Damage.
The regression equation is
% Damage = 84.1 - 52.5 Training - 44.8 Experience
- 18.5 Response to Alarm + 4.75 Age + 13.3
Qualification + 32.0 T*X - 21.5 G*Q
Predictor Coef SE Coef T P
Constant 84.125 3.793 22.18 0.000
Training -52.500 3.793 -13.84 0.000
Experience -44.750 3.793 -11.80 0.000
Response to Alarm -18.500 2.682 -6.90 0.000
Age 4.750 3.793 1.25 0.246
Qualification 13.250 3.793 3.49 0.008
T*X 32.000 5.365 5.96 0.000
G*Q -21.500 5.365 -4.01 0.004
S = 5.36482 R-Sq = 98.1% R-Sq(adj) = 96.4%
Table 7. Analysis of Variance for % Damage.
Analysis of Variance
Source DF SS MS F P
Regression 7 11659.5 1665.6 57.87 0.000
Residual Error 8 230.2 28.8
Total 15 11889.8
Source DF Seq SS
Training 1 5329.0
Experience 1 3306.3
Response to Alarm 1 1369.0
Age 1 144.0
Qualification 1 25.0
T*X 1 1024.0
G*Q 1 462.3
Alharthi A. A et al Int. Journal of Engineering Research and Applications www.ijera.com
ISSN : 2248-9622, Vol. 4, Issue 5( Version 2), May 2014, pp.37-44
www.ijera.com 43 | P a g e
The prediction equation is:
YDamage =84.1–52.5XT –44.8XX –18.5XR+ 4.75 XG+
13.3 XQ+32XT XX –21.5XGXQ ….. (Equation (3))
The residual plot (figure 11) indicates that
there is no violation of the analysis of variance
assumptions. The residuals are normally distributed,
the residuals have equal variances, and the residuals
are independent. This concludes that our model is
valid.
20100-10-20
99
90
50
10
1
Residual
Pe
rce
nt
25020015010050
20
10
0
-10
-20
Fitted ValueR
esi
du
al
20151050-5-10-15
4.8
3.6
2.4
1.2
0.0
Residual
Fre
qu
en
cy
16151413121110987654321
20
10
0
-10
-20
Observation Order
Re
sid
ua
l
Normal Probability Plot Versus Fits
Histogram Versus Order
Residual Plots for Extinguishing Time
Figure 10. Residual analysis for extinguishing time
1050-5-10
99
90
50
10
1
Residual
Pe
rce
nt
806040200
5
0
-5
-10
Fitted Value
Re
sid
ua
l
5.02.50.0-2.5-5.0-7.5-10.0
4.8
3.6
2.4
1.2
0.0
Residual
Fre
qu
en
cy
16151413121110987654321
5
0
-5
-10
Observation Order
Re
sid
ua
l
Normal Probability Plot Versus Fits
Histogram Versus Order
Residual Plots for % Damage
Figure 11. Residual analysis for % damage model
Alharthi A. A et al Int. Journal of Engineering Research and Applications www.ijera.com
ISSN : 2248-9622, Vol. 4, Issue 5( Version 2), May 2014, pp.37-44
www.ijera.com 44 | P a g e
4.3 Relation between the time of extinguishing
and the damage :
Correlations: Extinguishing Time,% Damage
Pearson correlation of Extiguishing Time and %
Damage = 0.945
P-Value = 0.000
260240220200180160140120100
90
80
70
60
50
40
30
20
10
0
Extiguishing Time
% D
am
ag
e
Scatterplot of % Damage vs Extiguishing Time
Figure 1 : Extinguishing Time vs Damage
Percentage
Figure 12 clarifies the relation between the
time of extinguishing and the damage percentage. It
is obvious that the damage percentage increase with
increasing the time. From the results in the last
section, it is concluded that the training, the
experience , the response to alarm and the interaction
between the training and the experience have the
most influence in decreasing the percentage of
damage, the age has small influence and the
qualification approximately has no influence.
V. Conclusions Our goal is to decrease extinguishing time
and percent damage, the factor levels should be set to
that produce the lowest mean. Examining the main
effects plots and interaction, the factor levels that
decrease extinguishing time and percent damage are
summarized in table below
Factors Level
Training 1
Experience 1
Response to Alarm 1
Age 1
Qualification 1
In conclusion:
a) The training has the highest effect in regard of
the performance of the employees in fire
extinguishing.
b) The experience factor is ranked in the second
stage according to its effect in the performance
of the employee in fire extinguishing.
c) The interaction between the training and the
experience was also a significant factor.
d) The age has small affect in the performance of
the employees in fire extinguishing.
e) The qualification’s effect on the performance of
the employees in fire extinguishing can also be
neglected.
A trained employee, less than 40 years of
age, with a bachelor degree and experience, and with
fast response leads to the best result in extinguishing
time and percent damage.
References [1] Bala MG, Biswanath M & Sukamal G,
Taguchi Method and ANOVA: An approach
for process parameters optimization of hard
machining while machining hardened steel,
Journal of Scientific & Industrial Research,
Vol. 68, August 2009, pp. 686-695
[2] Carey, W.M., National Class A Foam
Research Project Technical Report:
Knockdown, exposure and Retention Tests.
National Fire Protection Research
Foundation,1993
[3] Local Government Group (2011) Fire safety
in purpose-built blocks of flats,
http://www.local.gov.uk/c/document_library
/get_file?uuid=71e152a6-9e0a-4810-aee6-
498167664f79&groupId=10171
[4] Peace, G. S., Taguchi Methods: a Hands-on
Approach, Addison-Wesley Publishing
Company, Inc., Massachusetts, 1992.
[5] Ross, R. J., Taguchi Techniques for Quality
Engineering, McGraw-Hill, New York,
1989.
[6] Under writers Laboratories Inc., Report of
Class A Foam Tests, Department of the
Army, Fort Belvoir, 1994.